Mobile application for stress detection via MFCC speech-features

Everyone goes through periods of mental stress in their daily lives. While small amounts of stress can enhance focus, alertness, and performance, excessive stress negatively affects productivity, emotional stability, quality of life, and overall health. Early stress detection is crucial for preventing its harmful effects and managing it effectively. Despite significant research in stress detection, there remains a lack of real-time, mobile-based, and intuitive solutions that are easily accessible to the public. This research focuses on developing StressVox, a mobile application that uses speech analysis to identify stress or not stress. The system, developed using Flutter, utilizes a pretrained Convolutional Neural Network (CNN) trained on the Toronto Emotional Speech Set (TESS) dataset. The methodology includes data preparation, Mel Frequency Cepstral Coefficients (MFCC) feature extraction, noise handling, and speaker verification to ensure accurate and personalized results. The application uses a white and blue theme for a calming user experience and integrates with a Python backend via APIs, enabling real-time stress classification. The system achieved a 96% accuracy rate in detecting stress and includes a feedback-driven feature to rework the model when false results occur, further enhancing its performance over time. StressVox is designed to empower users with an accessible and non-invasive tool for early stress detection and management, providing a practical and user centric approach to addressing the growing need for mental health monitoring in daily life.